U.S. patent application number 13/534899 was filed with the patent office on 2014-01-02 for detection of defects embedded in noise for inspection in semiconductor manufacturing.
The applicant listed for this patent is Jason Z. LIN. Invention is credited to Jason Z. LIN.
Application Number | 20140002632 13/534899 |
Document ID | / |
Family ID | 49777740 |
Filed Date | 2014-01-02 |
United States Patent
Application |
20140002632 |
Kind Code |
A1 |
LIN; Jason Z. |
January 2, 2014 |
DETECTION OF DEFECTS EMBEDDED IN NOISE FOR INSPECTION IN
SEMICONDUCTOR MANUFACTURING
Abstract
One embodiment relates to an apparatus for detecting defects on
a manufactured substrate. The apparatus includes an imaging tool
arranged to obtain image frames from the manufactured substrate.
The apparatus further includes a data processing system which
includes computer-readable code configured to compute features for
pixels in an image frame and divide the pixels in the image frame
into feature-defined groups of pixels. The computer-readable code
is further configured to select a feature-defined group, and
generate a multi-dimensional feature distribution for the selected
feature-defined group. Another embodiment relates to a method of
detecting defects from a test images frame and multiple reference
image frames. Other embodiments, aspects, and features are also
disclosed.
Inventors: |
LIN; Jason Z.; (Saratoga,
CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
LIN; Jason Z. |
Saratoga |
CA |
US |
|
|
Family ID: |
49777740 |
Appl. No.: |
13/534899 |
Filed: |
June 27, 2012 |
Current U.S.
Class: |
348/87 ;
382/149 |
Current CPC
Class: |
G06T 7/001 20130101;
G06T 2207/30148 20130101; G06T 2207/20081 20130101; G01N 21/95607
20130101; G01N 2021/8887 20130101; G01N 2021/95615 20130101; G06T
2207/20076 20130101; G01N 21/8851 20130101 |
Class at
Publication: |
348/87 ;
382/149 |
International
Class: |
G06K 9/46 20060101
G06K009/46; H04N 7/18 20060101 H04N007/18 |
Claims
1. An apparatus for detecting defects on a manufactured substrate,
the apparatus comprising: an imaging tool arranged to obtain image
frames from the manufactured substrate; and a data processing
system including a processor, memory and computer-readable code in
said memory, the computer-readable code being configured to compute
features for pixels in an image frame; divide the pixels in the
image frame into feature-defined groups of pixels; select a
feature-defined group; and generate a multi-dimensional feature
distribution for the selected feature-defined group.
2. The apparatus of claim 1, wherein the computer-readable code is
further configured to determine a normal cluster in the
multi-dimensional feature distribution.
3. The apparatus of claim 2, wherein the computer-readable code is
further configured to detect outlier points which are outside the
normal cluster.
4. The apparatus of claim 3, wherein the computer-readable code is
further configured to locate defective pixels associated with the
outlier points.
5. The apparatus of claim 4, wherein the computer-readable code is
further configured to report the defective pixels.
6. The apparatus of claim 1, wherein the features include reference
features, and wherein a reference feature is a property associated
with a pixel location on multiple reference images.
7. The apparatus of claim 6, wherein the features further include
test features, and wherein a test feature is derived from a pixel
location on a test image and the multiple reference images.
8. The apparatus of claim 6, wherein the property comprises a range
of gray levels at the pixel location.
9. The apparatus of claim 6, wherein the property includes
information from a local range of pixels centered at the pixel
location.
10. A method of detecting defects from a test images frame and
multiple reference image frames, the method comprising: computing
features for pixels in the test and reference image frames;
dividing the pixels in said image frames into feature-defined
groups of pixels; selecting a feature-defined group; and generating
a multi-dimensional feature distribution for the selected
feature-defined group.
11. The method of claim 10 further comprising: determining a normal
cluster in the multi-dimensional feature distribution.
12. The method of claim 11 further comprising: detecting outlier
points which are outside the normal cluster.
13. The method of claim 12 further comprising: locating defective
pixels associated with the outlier points.
14. The method of claim 13 further comprising: flagging the
defective pixels.
15. The method of claim 10, wherein the features include reference
features, and wherein a reference feature is a property associated
with a pixel location on the multiple reference image frames.
16. The method of claim 15, wherein the features further include
test features, and wherein a test feature is derived from a pixel
location on a test image frame and the multiple reference image
frames.
17. The method of claim 15, wherein the property comprises a range
of gray levels at the pixel location.
18. The method of claim 15, wherein the property includes
information from a local range of pixels centered at the pixel
location.
19. A non-transitory tangible data storage medium storing
computer-readable code configured to perform a method comprising:
computing features for pixels in the test and reference image
frames; dividing the pixels in said image frames into
feature-defined groups of pixels; selecting a feature-defined
group; and generating a multi-dimensional feature distribution for
the selected feature-defined group.
20. The non-transitory tangible data storage medium of claim 19,
wherein the method further comprises: determining a normal cluster
in the multi-dimensional feature distribution; detecting outlier
points which are outside the normal cluster; locating defective
pixels associated with the outlier points; and flagging the
defective pixels.
Description
BACKGROUND
[0001] 1. Technical Field
[0002] The present invention relates generally to wafer and reticle
inspection apparatus and methods of using same.
[0003] 2. Description of the Background Art
[0004] Automated inspection and review systems are important in
process control and yield management for the semiconductor and
related microelectronics industries. Such systems include optical
and electron beam (e-beam) based systems.
[0005] In the manufacture of semiconductor devices, detection of
defects early on in the development and fabrication process is
becoming increasingly important to shorten product development
cycles and increase manufacturing yield. Advanced wafer and reticle
inspection systems are being used to detect, review and classify
defects and feed the root cause information back into the
manufacturing process to prevent these defects going forward. The
size of relevant defects is directly proportional to the design
rule being applied to the manufacturing of the semiconductor
devices. As the design rule being applied continues to shrink, the
performance demands on inspection systems increases both in terms
of imaging resolution and speed (defects processed per hour).
SUMMARY
[0006] One embodiment relates to a method of detecting defects from
a test images frame and multiple reference image frames. Features
for pixels in the test and reference image frames are computed, and
the pixels in said image frames are divided into feature-defined
groups of pixels. A feature-defined group is selected, and a
multi-dimensional feature distribution is generated for the
selected feature-defined group.
[0007] In addition, a normal cluster in the multi-dimensional
feature distribution may be determined, and outlier points which
are outside the normal cluster may be detected. Defective pixels
associated with the outlier points may be located, and the
defective pixels may be flagged and/or reported.
[0008] One embodiment relates to an apparatus for detecting defects
on a manufactured substrate. The apparatus includes an imaging tool
arranged to obtain image frames from the manufactured substrate.
The apparatus further includes a data processing system which
includes computer-readable code configured to perform a method of
detecting defects which divides pixels into feature-defined groups
and generates group-specific feature distributions in order to
detect outlier points.
[0009] Other embodiments, aspects, and features are also
disclosed.
BRIEF DESCRIPTION OF THE DRAWINGS
[0010] FIG. 1 is a flow chart depicting a method of detecting
defects in image data in accordance with an embodiment of the
invention.
[0011] FIG. 2 is a diagram depicting an example inspected area
within an image frame in accordance with an embodiment of the
invention.
[0012] FIGS. 3 and 4 show example two-dimensional feature
distribution plots which are generated from an image frame of an
inspected area that is similar to the framed area depicted in FIG.
2
[0013] FIG. 5 is a schematic diagram of an inspection apparatus
that may be utilized for automatically inspecting manufactured
substrates in accordance with an embodiment of the invention.
DETAILED DESCRIPTION
[0014] According to an embodiment of the present invention, images
from at least two dice may be used to perform the defect detection.
The image associated with the die in which defects are to be
detected may be referred to as the test image, and the images
associated with other dice may be referred to as the reference
images.
[0015] The present application discloses an innovative technique to
improve sensitivity of defect detection during the inspection of
manufactured semiconductor substrates. In particular, the technique
improves the sensitivity of detecting defects which may be
"embedded in noise" in a feature distribution plot
(multi-dimensional signal) over a multi-dimensional feature
space.
[0016] To form the feature distribution plot, each point in the
multi-dimensional feature space may be assigned with a population
value. For example, if there are one hundred pixels in the frame
area having identical test and reference features, the population
value of the feature point is assigned as one hundred.
[0017] In the technique disclosed previously in U.S. Pat. No.
7,440,607, the feature distribution plot is preferably formed based
on the features of the pixels within the area of a frame, as
derived from the test image and the multiple reference images.
Because the values of the multiple features of non-defective
(normal) pixels fall within certain nominal ranges, a normal
cluster (normal distribution) is generally formed by the feature
distribution of non-defective pixels in the feature distribution
plot. Defective pixels become outliers in the feature distribution
plot because one or more of their feature values do not fall in the
nominal ranges.
[0018] In order to improve the defect detection, more features may
be used in generating the feature distribution plot. However,
applicant has determined that, when more than two dimensions are
used, the normal cluster (normal distribution) formed by the
non-defective pixels in the feature distribution plot often becomes
sparse, making it difficult to separate outliers from the normal
cluster.
[0019] In contrast, in accordance with an embodiment of the present
invention, pixels of a frame are divided into separate
feature-defined groups based on one or more reference features.
Subsequently, one of the groups of pixels may be selected, and the
feature distribution plot may be formed based on the test and
reference features of pixels within the selected group (without
including pixels outside the group).
[0020] Applicant has determined that this technique is advantageous
in that it allows for the detection of defects which are not
detected previously. This is because the previously un-detected
defects were embedded in the "noise" around the normal cluster
based upon all pixels in a frame. In contrast, these defects become
detectable outliers when the normal cluster is based upon pixels in
a selected feature-defined group. This is because when a separate
distribution plot is created for a separate group of pixels, the
noise attributable to the non-selected groups is effectively
removed from the plot.
[0021] FIG. 1 is a flow chart depicting a method 100 of detecting
defects in image data in accordance with an embodiment of the
invention. The image data may include a test image and
corresponding reference images. The reference and test images may
be each divided into a number of frames. Each frame may cover a
significant number of pixels in a local area of a die. The frame
may be rectangular or square. For instance, 512.times.512 or
1024.times.1024 pixels may form a square frame.
Feature Computation
[0022] The first step 102 of the method 100 involves the
computation of reference features and test features for pixels in
reference and test images. The reference and test features may be
determined as follows.
[0023] A reference feature may be defined as some property
associated with a same (corresponding) pixel location on multiple
reference images. For example, the reference feature may be the
average or median of gray levels across multiple reference dice at
the same pixel location. As another example, a reference feature
may be the range or deviation of the gray levels across multiple
reference dice at the pixel location.
[0024] A reference feature derived from a pixel location may also
include the information around the pixel location. For example, the
local range or local average of a
three-pixel-wide-by-three-pixel-high area centered at the pixel
location (or other local area defined relative to the pixel
location) on each reference image may be calculated first, and then
these local ranges or averages may be used to derive the reference
feature of the pixel location. For instance, a median of local
averages across multiple dice may be one of the reference features
of the pixel location. Many other features may be derived according
to these principles.
[0025] A test feature may be derived from a pixel location on the
test image as well as the multiple reference images. One example of
a test feature is the difference between the gray level on the test
image and the average of the gray levels on the reference images.
Another example of a test feature is the difference between the
gray level on the test image and the median of the gray levels on
the reference images. Similar to a reference feature, a test
feature derived from a pixel location may also include the
information around the pixel location. For example, a local average
around a three-pixel-wide-by-three-pixel-high area centered at the
pixel location (or other local area defined relative to the pixel
location) on the test image may be used for computing the
difference to the average of the local averages around a
three-pixel-wide-by-three-pixel-high area of the corresponding
location on the multiple reference images.
Feature-Defined Grouping and Group-Specific Feature
Distributions
[0026] In the second step 104, the pixels of a frame may be divided
into feature-defined groups. In this step, at least one feature is
used to separate at least one group of pixels from other pixels in
the inspected area. For example, pixels of a densely patterned area
typically have more local gray level range variation than pixels in
an open area. Therefore, a local gray level range may be used as a
feature for dividing pixels into different population groups.
[0027] In the third step 106, a multi-dimensional feature
distribution may be generated for one or more selected groups of
pixels. In contrast, a previous technique would use all pixels of
an image frame to generate a multi-dimensional feature distribution
to be used for outlier detection.
[0028] For example, consider the inspected area as shown in the
image frame 200 depicted in FIG. 2. As shown, the image frame 200
includes vertical lines 202, horizontal line segments 204, and open
spaces 206.
[0029] A previous technique would use all pixels in the frame 200
to generate a multi-dimensional feature distribution. In contrast,
the presently-disclosed technique first divides the pixels in the
frame 200 into groups based on a computed feature per the second
step 104.
[0030] In a first example, using a local gray level range or other
computed feature, the pixels of the frame 200 may be separated into
a first group including the pixels of both the vertical lines 202
and horizontal line segments 204, and a second group including the
pixels of the open spaces 206. In a second example, the pixels of
the frame 200 may be separated into a first group including the
pixels of the vertical lines 202, a second group including the
horizontal line segments 204, and a third group including the
pixels of the open spaces 206.
[0031] Subsequently, per the third step 106, one or more of the
feature-defined groups may be selected, and a multi-dimensional
feature distribution may be generated separately for each of the
selected groups.
Normal Cluster Determination
[0032] The fourth step 108 of the method 100 may determine the
pixels that fall into the normal cluster (normal distribution) of
points for each group-specific feature distribution. In accordance
with embodiments of the present invention, there may be several
valid approaches for defining the normal cluster of pixels.
[0033] One approach to identify the normal distribution is based on
the population in the local neighborhood of a point location of the
signal distribution formed in the multi-dimensional feature space.
For a two-dimensional example, a pre-defined population density
threshold for the total population of points within, for example, a
five-pixel-wide-by-five-pixel-high square area centered on a given
point in the two-dimensional signal distribution may be used as a
threshold to determine if the given point is normal or not. If the
population value is greater than the population density threshold,
the point may be considered to be normal.
[0034] Another approach for identifying the normal distribution is
based on the connectedness among points in the signal distribution.
For a two-dimensional example, a point may be considered normal if,
in the signal distribution, there are other points within a
pre-defined distance.
[0035] Other approaches may also be used to define the normal
distribution. As an example, a point may be considered normal if it
satisfies both criteria described above. In principle, if there are
a considerable number of pixels that have identical or similar
features, they are considered normal and not defective.
Outlier Detection
[0036] The fifth step 110 may identify the pixels that are
statistical outliers, which may be indicative of defects on the
test substrate. As described above, the normal cluster of points in
the group-specific feature distribution has been identified. Every
point that is not identified as within the normal cluster may be
considered as a candidate point. A candidate point may contain one
or more pixels that correspond to one or more real defects.
[0037] In order to allow some margin for error, a tolerance range
may be allowed before a point is declared defective. There may be
different approaches in defining the tolerance range. One approach
is a fixed tolerance range. In a two dimensional example, the
tolerance range may be a pre-defined distance from a normal point
to the candidate point. If the candidate point is disposed at a
greater distant from any normal point than the pre-defined
distance, it may be declared a defective point in that pixels
associated with the candidate point may be deemed to be defective.
More complex rules such as making the pre-defined distance a
function of the reference features may also be added in defining
the tolerance range. As an example, the distance may be a function
of a reference feature that is defined as the average gray level
across multiple reference dice.
[0038] To achieve optimal performance in identifying real defects
and eliminating nuisances, parameters such as the tolerance range
are tuned. Each defective point may be viewed in the
multi-dimensional feature space, and the tolerance range may be
tuned to capture points corresponding to defects of interest, while
the points corresponding to nuisances are eliminated.
Defective Pixel Identification
[0039] As described above, outlier points may be detected in a
group-specific feature distribution during the fifth step 110, and
pixels associated with the outlier points may be deemed to be
defective. It is noted that in the multi-dimensional feature space,
each point represents one or more pixels that have certain feature
characteristics. However, the actual pixel locations where the
pixels are in the frame area may not be directly retained in the
feature distribution. Therefore, it may be necessary to perform
additional processing to locate these defective pixels per the
sixth step 112.
[0040] Once the defective pixels associated with the outlier points
have been located, these defective pixels may be flagged and
reported per the seventh step 114. In one embodiment, the report
may be made to the tool operator and/or to an engineering database,
so that an investigation of the defects may be made. In some
embodiments, the tool in which the present method of defect
detection is implemented also includes methods for defect analysis
and identification. In other embodiments, the two functions of
detection and analysis are separately implemented.
Example Feature Distribution Plots
[0041] FIGS. 3 and 4 show example two-dimensional feature
distribution plots which are generated from an image frame of an
inspected area that is similar to the framed area 200 depicted in
FIG. 2 in that it has vertical lines 202, horizontal line segments
204, and open areas 206. A darker point in the plots indicates a
greater population of pixels with feature values in the ranges
associated with that point.
[0042] The feature distribution plot 300 in FIG. 3 is generated
from all pixels in the image frame 200. The normal cluster 302 is
clearly visible in the plot. In this case, there is a point 304
which has associated with it pixels which are actually defective.
However, in this feature distribution 300, the point 304 falls
within the "noise" near the edge of the normal cluster 302. Hence,
the point 304 is not identifiable as an outlier in this
distribution.
[0043] In contrast, the feature distribution plot 400 in FIG. 4 is
generated from only pixels of the vertical lines 202 in the image
frame 200. The normal cluster 402 is again clearly visible in the
plot. The normal cluster 402 in FIG. 4 is substantially smaller
(tighter) than the normal cluster 302 in FIG. 3. In this case, the
point 404 which has associated with it pixels which are actually
defective is outside the normal cluster 402. Hence, in this feature
distribution 400, the point 404 is identifiable as an outlier.
High-Level Diagram of Apparatus
[0044] FIG. 5 is a schematic diagram of an inspection apparatus
that may be utilized for inspection of manufactured substrates in
accordance with an embodiment of the invention. As shown in FIG. 5,
the inspection apparatus includes an imaging tool 510, a movable
substrate holder 520, a detector 530, a data processing system 540,
and a system controller 550.
[0045] In one embodiment, the imaging tool 310 comprises an
electron beam (e-beam) imaging column. In an alternate embodiment,
the imaging tool 510 comprises an optical imaging apparatus. In
accordance with an embodiment of the invention, the imaging tool
510 includes electronics to control and adjust the magnification of
the imaging.
[0046] The movable substrate holder 520 may comprise a translatable
mechanism to hold a target substrate 525. The target substrate 525
may be, for example, a semiconductor wafer or a reticle for
lithography. The detector 530 is an appropriate detector for the
particular imaging apparatus, and the data processing system 540 is
configured to process image data from the detector 530. The data
processing system 540 may also include a processor 542, memory 544
for holding computer-readable code 545, a data storage system 548
for storing data, and various other components, such as a system
bus, input/output interfaces, and so forth.
[0047] The system controller 550 may be communicatively coupled to
the imaging tool 510 so as to electronically control the operation
of the imaging tool 510. The system controller 550 may also include
a processor 552, memory 554 for holding computer-readable code 555,
a data storage system 556 for storing data, and various other
components, such as a system bus, input/output interfaces, and so
forth.
Conclusion
[0048] In the above description, numerous specific details are
given to provide a thorough understanding of embodiments of the
invention. However, the above description of illustrated
embodiments of the invention is not intended to be exhaustive or to
limit the invention to the precise forms disclosed. One skilled in
the relevant art will recognize that the invention can be practiced
without one or more of the specific details, or with other methods,
components, etc.
[0049] In other instances, well-known structures or operations are
not shown or described in detail to avoid obscuring aspects of the
invention. While specific embodiments of, and examples for, the
invention are described herein for illustrative purposes, various
equivalent modifications are possible within the scope of the
invention, as those skilled in the relevant art will recognize.
[0050] These modifications can be made to the invention in light of
the above detailed description. The terms used in the following
claims should not be construed to limit the invention to the
specific embodiments disclosed in the specification and the claims.
Rather, the scope of the invention is to be determined by the
following claims, which are to be construed in accordance with
established doctrines of claim interpretation.
* * * * *